BBW alpha — interactive pareto

35 arms · 86 tasks · pi 0.85.1 · click a point for full config · hover to preview · frontier is drawn within each VRAM envelope

UnbiasedStrata R2 passed the full task, engine and log audits and participates in the frontier. Its serving engine, INT8 KV, 192k context and uncapped reasoning differ from the llama-server runs; see the config and R2 task audit. The earlier R1 remains provisional because of serving and harness defects and does not set the frontier. R1 audit.

0.040.230.420.600.790.98246810121416UnbiasedStrata Flash-Next IQ3_XXS +MTPMuse Spark Open 816B-A42B IQ3_XXS (2× B200)Qwen3.6-27B IQ4_XSUnbiasedStrata Flash-Next IQ3_XXS +MTP (rotated Q4 KV)UnbiasedStrata Coder IQ1_M +MTP (32 GiB RAM)UnbiasedStrata Flash-Next IQ3_XXS (provisional)ThinkingCap-3.6-27B Q4_K_M +MTPMuse-Glimmer-30B IQ3_XXSThinkingCap-3.8-27B Q4_K_M +MTP (medium)Qwen3.8-27B GSQ-RCO IQ3_S +MTPQwen3.6-35B-A3B IQ4_XSQwen3.6-35B-A3B Q4_K_M +MTP 256k expert-LRUQwen3.8-27B Q3_K_MQwen3.6-35B-A3B IQ3_XXS (8GB)Saluki-27B GSQ-RCO IQ2-mixQwen3.6-35B-A3B REAP-48 (8GB)Gemma-4-26B-A4B Q4_K_MQwen3.6-35B-A3B IQ2_M (8GB)Bonsai-2-27B PTQ1_0 256k pen0Heretic 35B-A3B Q3_K_M 256kBonsai-2-27B PTQ1_0 256kBonsai-2-27B PTQ1_0 48k (8GB)Qwen3.5-9B Q4_K_M (8GB)HauhauCS 35B-A3B IQ2_MGLM-4.7-Flash Q4_K_M 256kQwen3.5-9B Q4_K_MGemma-4-26B-A4B IQ3_XXS (8GB)HauhauCS 35B-A3B IQ2_M (8GB)Xing4.0-29B-A4B IQ4_NL +MTPGemma-4-12B IQ4_XS 40k (8GB)Bonsai-27B Q1_0 128k (8GB)Granite-4.2-8B Q4_K_M 48k (8GB)Qwen3.8-27B IQ2_XXS 16k (8GB)Huihui 35B-A3B Q2_KQwen3.8-27B UD-IQ1_S 256k wall-clock hours to work the suite (lower is better) mean score (higher is better)

No arm selected

Click any point to pin its configuration here. Hover to preview without pinning.
view:
8 GB envelope (11) 16 GB envelope (8) 24 GB envelope (15) >24 GB / multi-GPU envelope (1) on its envelope's frontier dominated within its envelope
arms click to show/hide · frontiers recompute over what is visible